Spin-lock is commonly used for process serialization, where it works well for multi processor systems. Under some conditions however, it may cause an unexpected increase of CPU overhead. To address this problem a simulator has been developed which evaluates various algorithms of the lock/unlock process to determine a method to minimize their affect on the stability and scalability of a system. This paper analyzes the effect of spin-lock, how CPU overhead changes as a function of traffic, by using the simulator. It demonstrates how multiple processors go into ldquobusy waitrdquo, consuming CPU time, and working for nothing, with only a few processors able to advance. The paper also shows how this problem can be solved by capped spin-lock where the spin is capped to a certain limit and a pause is inserted between the spins avoiding unnecessary consumption of CPU power and maintaining scalability over a number of processors.

Dynamic changes in open distributed environments damage the continuous execution of distributed objects. A distributed environment consisting of reconfigurable distributed objects would have higher reliability and availability, and a greater resilience to unexpected requirement changes. In this paper, we propose a reconfigurable, object model that dynamically changes the object's behavior to fit the current environment by modifying its internal structure. The proposed object consists of communicating concurrent meta-objects. Each meta-object contains functions for adaptation, remote communication, and administration. Therefore, this model provides the required separation of concerns (SoC). Generative communication, which allows meta-object communication, gives the proposed reconfigurable objects the following characteristics: flexibility; ability of allowing a variety of configurations; safety that ensures consistency; and a unification of state preservation and communication. The proposed object model was successfully implemented in a middleware framework called Juice 2.

In distributed object-oriented computing, a complex set of requirements has to be considered for maintaining the required level of reliability of objects to give higher level of performance. The authors proposed a mechanism to analyze the complexity of these underlying environments and design a dynamically reconfigurable architecture according to the changes of the underlying environment. The replication should have a suitable architecture for the adaptable fault tolerance so that it can handle the different situations of the underlying system before the fault occurs. The system can provide the required level of reliability by measuring the reliability of the underlying environment and then either adjusting the replication degree or migrating the object appropriately. It is also possible to improve the reliability with shifting to a suitable replication protocol adaptively. However, there should be a mechanism to overcome the problems when "replication protocol transition period" exists. This will also find a client-server group communication protocol that communicates with objects under both different environmental conditions and different replication protocols.

Caching has long been used in most fields of the computer systems to enhance the scalability of the objects, improve the performance and reduce the access latency. A significant effort has been made to introduce cache-coherent algorithms for maintaining the consistency of such data objects in cache by keeping a higher freshness of the data. Updating the cache objects considering the access behavior and user preferences is one of an attractive solutions to maintain the consistency. In this paper, we define quality of data (QoD) metric to evaluate the amount of freshness that is necessary to satisfy the user requirements. We then focus on the update scheduling method that analyzes the access behavior of the cache objects and predicts the time interval for updating the cache. Here, we introduce the "average update interval method" that uses the most recent time between access values, to predict the time interval. Using our proposed algorithm, the user can not only access the preference view but also he can get the maximum QoD of the objects. Moreover we performed extensive experiments using web log data and simulation data. Then the results could conclude that the cache objects are maintaining more than 70% of consistency with the original objects.

We propose an access control model that is suitable for a distributed object oriented environment. Our model has two features: authentication with object properties and method categorization by a security level. The object property is meta information of a client, and the client is vested with it in advance. To use the object properties, a server can identify a huge number of clients in the environment by groups that is categorized with the object properties of the clients. And also, to use a combination of multiple object properties in authentication, an administrator of the server can determine the flexible range of target clients. The security level shows how much impact the method affects server's data. If a designer of the server categorizes the server's methods with a security level, an administrator of the server can set an authorization rule to each category instead of each server's method. The categories constitute a tree structure, since a parent category includes child categories. As a result of this, the administrator can set an authorization rule easier than authorization without categorization by the security level. Furthermore, we implemented above the access control model and we evaluated their efficiency.

New applications are being introduced for different areas such as commercial, education, and medical etc. A significant effort has been made in developing some of these applications considering the adaptability to the dynamically changing network and hosts' environments. In client-server applications, this adaptability is very important to enhance the reliability of the system and maintain the consistency of the objects. We describe a framework that can detect the changes in performance of the execution environment, and provide strategies in order to adapt such changes. Using these access strategies the client objects can access the server objects efficiently in a dynamically changing environment. We do not only consider the dynamic changes of the network characteristics but also consider the end-hosts performances before changing to the optimal strategy. Therefore the client objects can achieve the maximum usefulness of the environment. Here, we introduce three access strategies, namely client migration, direct access, and lazy access to impart an adaptive behavior to the client objects in the distributed environment. Advancing from the conventional static access models we propose algorithms to select the best-suited access strategy for the dynamic environment. Moreover the proposed model is to be implemented to the "juice" distributed object-oriented environment.

To enhance reliability of distributed systems, replication techniques are used. We propose an adaptable replication scheme that enables an object to adapt itself to provide a replicated service, and a replicated object to choose the replication strategy that best fits to achieve its requirements. What kind of replication strategy should be used depends on an execution environment as well as an application. The adaptable replication scheme permits replacement of down replicas, or change of the number of replicas when partial failures occur and chooses the most adequate consistency protocol for the current configuration. Clients to the recently adapted objects for replication should take actions that enables then to obtain a thorough service of the replicated object. The adaptable replication scheme is being designed as part of the Juice system. The Juice object model allows objects to change its behavior on the fly by replacing some components in a modular way.

This paper presents a language framework for a concurrent description and an implementation of a synchronization mechanism in distributed object-oriented computation. Some concurrent object-oriented languages are considered to be constructed from a user-defined function and a synchronization constraint. We propose the language for the synchronization constraint, which allows to define mutual exclusion and conditional synchronization by using declarative notations. The synchronization constraint is separated from user-defined functions, and the relation between them is described in the binding code. The binding mechanism reduces the dependency between them and contributes to code reuse. This framework makes it is easy for programmers to understand and describe the use-defined functions and synchronization constraints independently. The implementation of the synchronization mechanism is based on a reconfigurable object model that we are developing. The user-defined function and synchronization mechanism are implemented as its own separated module in the single object. Then the synchronization mechanism has flexibility and preserves the concept of object-orientation.

Reinforcement learning is a kind of machine learning that adapts to an environment with special input called a reinforcement signal. An agent using reinforcement learning can obtain purposeful behavior autonomically. However, there are problems in that reinforcement learning takes a long time because it advances while repeating trial-and-error, and an acquired action is not necessarily optimal. We propose reinforcement learning using state transition rates, and compare it with another method. As a result, our method shows the capability of learning purposeful behavior efficiently.

In analyzing electronic circuits, it is usually necessary to solve simultaneous linear equations which provide with a sparse coefficient matrix. In order to treat these problems effectively, we propose a dedicated parallel machine called 'Sparse Matrix Solving Machine', or (SM)2 for short.